Research map / Vision-language and multimodal / Visual question answering
Towards Models that Can See and Read
Roy Ganz, Oren Nuriel, Aviad Aberdam, Yair Kittenplon, Shai Mazor, Ron Litman
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add OCR tokens to the existing text encoder naively (ablation) | BLIP · TextVQA val (Comb.) | accuracy: 32.43 -> 43.02 | Improved |
| add designated OCR encoder with 2-D info (ablation) | UniTNT_BLIP · TextVQA val (Comb.) | accuracy: 43.02 -> 46.13 | Improved |
| add VL-OCR decoding fusion (ablation) | UniTNT_BLIP · TextVQA val (Comb.) | accuracy: 46.13 -> 47.38 | Improved |
| add OCR language-modeling loss L_OCR-LM (ablation) | UniTNT_BLIP · TextVQA val (Comb.) | accuracy: 47.38 -> 52.66 | Improved |
| add OCR-BC loss with 2-D information (ablation) | UniTNT_BLIP · TextVQA val (Comb.) | accuracy: 52.66 -> 55.21 | Improved |
| freeze the vision encoder (ablation) | UniTNT_BLIP · TextVQA val | accuracy: 52.45 -> 55.21 | Improved |
| add scene-text understanding (UniTNT) with VQAv2-only training | BLIP · VQAv2 and TextVQA | accuracy: +3.19% VQA, +12.16% TextVQA | Improved |
| add scene-text understanding (UniTNT) with TextVQA-only training | BLIP · VQAv2 and TextVQA | accuracy: +19.67% TextVQA, -3.15% VQA | Mixed |
| add UniTNT with TextCaps-only finetuning | BLIP · TextCaps vs COCO Captions | CIDEr: +19.4 TextCaps test, -14.4 COCO | Mixed |
| combined training on vision and scene-text data rather than separate finetuning (ablation) | UniTNT_BLIP · VQA | average accuracy: 58.01 -> 67.56 | Improved |
| combined training instead of TextCaps-only finetuning (ablation) | UniTNT_BLIP · TextCaps val | CIDEr: 130.5 -> 119.1 | Hurt |
| add UniTNT to BLIP in Caps-only training | BLIP · COCO Karpathy-test | CIDEr: 133.3 -> 133.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BLIP — BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (ICML 2022) · better on 8 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Question Aware Vision Transformer for Multimodal Reasoning (CVPR 2024 · similarity 0.89)
- VisionLLM v2: An End-to-End Generalist Multimodal Large Language Model for Hundreds of Vision-Language Tasks (NeurIPS 2024 · similarity 0.89)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.88)
- BLIVA: A Simple Multimodal LLM for Better Handling of Text-Rich Visual Questions (AAAI 2024 · similarity 0.88)
- JourneyBench: A Challenging One-Stop Vision-Language Understanding Benchmark of Generated Images (NeurIPS 2024 · similarity 0.88)
- ClawMachine: Learning to Fetch Visual Tokens for Referential Comprehension (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
